test(tutorials): poll task state instead of a fixed 1s sleep

[danielmillerp]

What

The async base tutorials 010_multiturn and 020_streaming now poll the task state instead of sleeping a fixed second before asserting it.

Why

Each test sends an event, waits for the agent's reply, sleeps 1s, then asserts the task state holds 3 messages. The agent sends its reply first and persists the turn to state afterwards. When the model call is slow, the 1s read sees only the system message and the test fails with assert 1 == 3. This shows up as random red Agent Integration Tests on unrelated server PRs.

How

• test_utils/async_utils.py: new wait_for_state_messages, which polls the state until it reaches the expected count or a 30s timeout, then returns the last read.
• Both tutorials use it in place of the sleep-then-read blocks (4 call sites).

A state that never updates still fails, with its real length. The initial-state checks after task creation are unchanged.

Verification

• Unit tests (tests/lib/test_tutorial_async_utils.py, new, 3 passed) against a fake client:
  • state written 0.3s after the reply → waits and returns 3 messages (a fixed sleep shorter than the write delay reads 1)
  • state never written → returns 1 message at the 0.3s deadline, so the caller's assert fails as before
  • state already written → returns on the first poll with no added delay
• ruff check and pyright on the changed files pass. All 52 CI checks passed on the first push.
• Not yet verified: a live run. The server repo's integration workflow runs the tests baked into the published tutorial images, so this change reaches it only after those images are rebuilt from this branch.

🤖 Generated with Claude Code

Confidence Score: 5/5

The PR appears safe to merge; both earlier findings are addressed and no new issue was found in this PR’s changes.

Findings

1.  Busy workers fail this test ▶

Fix with agent prompt

### Issue 1
tests/lib/test_tutorial_async_utils.py:undefined-67
If a busy worker wakes more than two seconds after this test starts, the wall-clock check fails even though `wait_for_state_messages` honored its deadline. The root suite runs tests in parallel, so this can add an unrelated red test to the suite this PR aims to make more reliable. Check the deadline without requiring a fast wake-up.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

The 010_multiturn and 020_streaming tutorial tests now poll task state before checking for three messages, since an agent reply can arrive before the turn is saved.

• Async tutorial checks wait for task messages to be saved.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Agent replies] --> B[Poll task state]
  B --> C{Three messages yet?}
  C -- Yes --> D[Check messages]
  C -- No, time remains --> B
  C -- No, deadline passed --> D

Loading

_{Reviews (5) · Last reviewed commit: "chore(tutorials): drop the unrelated uv...." · Reviewed by Greptile}

[greptile-apps]

Busy workers fail this test

If a busy worker wakes more than two seconds after this test starts, the wall-clock check fails even though wait_for_state_messages honored its deadline. The root suite runs tests in parallel, so this can add an unrelated red test to the suite this PR aims to make more reliable. Check the deadline without requiring a fast wake-up.

Prompt To Fix With AI

This is a comment left during a code review.
Path: tests/lib/test_tutorial_async_utils.py
Line: 67

Comment:
**Busy workers fail this test**

If a busy worker wakes more than two seconds after this test starts, the wall-clock check fails even though `wait_for_state_messages` honored its deadline. The root suite runs tests in parallel, so this can add an unrelated red test to the suite this PR aims to make more reliable. Check the deadline without requiring a fast wake-up.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[danielmillerp]

Fixed. The upper wall-clock bound is gone. The test now asserts only that the helper kept polling until the deadline (elapsed >= timeout) and returned the 1-message state. A late wake on a busy worker can no longer fail it.